Moore's Law
Moore's law is the observation, made by Gordon Moore in 1965, that the number of components that can be economically placed on an integrated circuit doubles at a regular interval — originally about every year, later restated as about every two years. It is an empirical trend about the economics and manufacturing of chips, not a law of nature, and it has served for roughly six decades as the pacing metronome of the semiconductor industry.
The source is a short article in the 19 April 1965 issue of Electronics magazine, titled Cramming More Components onto Integrated Circuits. Moore, then director of research at Fairchild Semiconductor, plotted the number of components on the cheapest chips against time from 1959 onward and noticed that the count had been roughly doubling every year. He extrapolated that to about 65,000 components per chip by 1975, a bold claim at a time when typical chips held only dozens of parts. Crucially, his point was economic: for any year there was a number of components that minimized cost per component, and that optimum was rising rapidly. In 1975, at an industry conference, he revised the pace to a doubling about every two years, and the name Moore's law, attributed to Caltech professor Carver Mead, became universal. Moore later co-founded Intel with Robert Noyce in 1968.
For decades the trend held because of a second, quieter rule called Dennard scaling. In 1974 Robert Dennard and his IBM colleagues described how shrinking a MOSFET's dimensions and voltage together lets it switch faster while keeping power density roughly constant. This meant every new generation of smaller transistors was a pure win: more transistors, running faster, with the same power per unit area. Combined with steady advances in lithography, which prints the patterns on silicon, it let Intel's 4004 microprocessor of 1971, with about 2,300 transistors, evolve into chips with tens of billions of transistors, such as the processors in modern phones and the large AI accelerators with many tens of billions.
Dennard scaling broke down around the mid-2000s, and that is the change people most often feel without naming. As transistors became very small, leakage current rose, and supply voltages could not drop proportionally, so packing more transistors together generated too much heat to run them all at ever higher clock speeds. Processor frequencies, which had risen for decades, plateaued at a few gigahertz, and designers shifted to multiple cores, wider parallelism, and specialized hardware. This is why a modern computer is not simply faster by doubling each cycle; the gains come from doing more in parallel, which is also the logic of graphics processors and AI chips.
Moore's law is slowing, but what that means is easy to misread. Each new process generation now costs far more to design and manufacture, requires extreme ultraviolet lithography and new transistor structures such as FinFETs and gate-all-around nanosheets, and delivers smaller gains in cost and power. The names of the process nodes, such as 3 nanometers, are marketing labels and no longer measure any actual feature on the chip. The industry response is to stack and combine chips rather than only shrink them, using chiplets and advanced packaging, and to design circuits specialized for a task, so that useful capability keeps rising even as the classic doubling cadence stretches.
A few pitfalls are worth naming. It is not a law of physics, and it is not a guarantee: it is a self-fulfilling roadmap that firms coordinated around, and it required enormous investment to maintain. It does not say that computers double in speed every two years, only that the count of transistors per chip tends to; and the statement that it is dead is as inaccurate as the statement that it is unchanged. Moore himself, who lived from 1929 to 2023, was cautious about extending the prediction indefinitely, and acknowledged that exponential trends eventually meet physical and economic limits, with atoms setting the ultimate floor on size.
Read Moore's law as a self-sustaining industrial schedule rather than a law of nature: it began as one engineer's observation about the economics of cramming components onto a chip, was kept alive for decades by shared targets and Dennard scaling, and now survives in a looser form as the industry searches for gains beyond simply making transistors smaller.